Nvidia’s GR00T 2 trains robots by watching humans, hits 95% success
Nvidia’s new GR00T 2 AI model trains robots by watching 100,000 hours of human video, achieving a 95% success rate on delicate tasks after just five demonstrations. Foxconn and Siemens are already testing it for AI server assembly and factory shifts. Alongside it, Nvidia released Isaac Lab 3.0, an open-source simulator that trains robots up to 2.5 times faster, and DLSS 4.5, which adds three presets to simplify upscaling for any GPU.
Editor, Lazyfounder
30 SEC SUMMARY
- Nvidia’s GR00T 2 AI model learns delicate hand movements by studying 100,000 hours of human video, hitting 95% task success after five demos.
- Foxconn, Siemens, and Nexuni are testing GR00T 2 for tasks like assembling AI servers, factory shifts, and laundry.
- Isaac Lab 3.0, Nvidia’s open-source robot-training simulator, cuts training time by up to 2.5× in early tests.
- DLSS 4.5 introduces three presets—K, L, and M—to tailor upscaling and frame generation for high-end, mid-range, and older GPUs.
- Robotics professor Ken Goldberg compares the shift to "the most exciting technological development since the internet in 1995."
TABLE OF CONTENTS
- GR00T 2 learns human movements from 100,000 hours of video
- Foxconn, Siemens, and Nexuni test GR00T 2 in factories and homes
- Isaac Lab 3.0 slashes robot training time
- DLSS 4.5 adds presets for every GPU tier
- 4K and path-tracing get dedicated DLSS presets
- Robotics expert hails GR00T 2 as a breakthrough
- What this means
- Key takeaways
- Sources
GR00T 2 learns human movements from 100,000 hours of video
Nvidia’s open AI model, GR00T 2, watches 100,000 hours of human video to teach robots delicate hand skills. After five demonstrations of a task, the model completes it correctly 95% of the time, up from 71% after a single demo, according to The Next Web. Nvidia also released the same 100,000 hours of first-person footage for anyone to use in training their own robots.
Foxconn, Siemens, and Nexuni test GR00T 2 in factories and homes
Foxconn is testing GR00T 2 to assemble AI servers, Siemens ran an Nvidia-powered humanoid for a full eight-hour factory shift, and Nexuni is experimenting with it for laundry folding. None of these companies have yet committed to production rollouts, but earlier versions of GR00T already drive LG’s humanoid robot slated for 2027.
Isaac Lab 3.0 slashes robot training time
Nvidia shipped Isaac Lab 3.0, an open-source simulator that trains robots in virtual environments. In one test, it cut training time by up to 2.5× compared to older versions. Walden Robotics, which raised $300 million in July, is already using the tool for its robotics projects.
DLSS 4.5 adds presets for every GPU tier
DLSS 4.5 arrives with three presets—K, L, and M—that let gamers pick the balance between performance and image quality without tweaking menus. Preset K, designed for high-end PCs, delivers near-native 4K visuals. Preset M fits mid-range cards like the RTX 4070, while Preset L keeps older GPUs like the RTX 30 or 20 Series running smoothly in ray-tracing games, according to Engadget.
4K and path-tracing get dedicated DLSS presets
For 4K setups, preset K is the go-to choice, offering the sharpest upscaling. Deep Learning Anti-Aliasing (DLAA) remains the gold standard for smoothing jagged edges, but it demands a powerhouse GPU like the upcoming RTX 50 Series. DLSS 4.5 also debuts Multi Frame Generation, which can double frame rates in path-traced games—preset B reduces visual glitches compared to preset A, though not all titles support it.
Robotics expert hails GR00T 2 as a breakthrough
Ken Goldberg, a UC Berkeley robotics professor who partnered with Nvidia on two projects, calls agentic robotics “the most exciting technological development since the internet in 1995.” The comment reflects the broader excitement around robots that can learn and adapt to tasks in real time.
What this means
Lazyfounder analysis — our interpretation, not reported fact.
GR00T 2 turns lab demos into pilot projects at factories and assembly lines—if the 95% success rate holds in noisy, real-world settings, deployment could accelerate faster than industrial robotics did in the 2010s.
Key takeaways
- GR00T 2’s 95% success rate after five demos suggests one-shot learning is no longer the bottleneck for delicate tasks like picking or assembling.
- Simulation training in Isaac Lab 3.0 now runs up to 2.5× faster, letting startups iterate on robot policies without buying more hardware.
- DLSS 4.5 presets let gamers trade GPU power for visual fidelity without digging into menus—preset K is the new default for 4K rigs.
Related on Lazyfounder
Sources
- The Next Web · 2026-11-09
Nvidia’s GR00T 2 teaches robots delicate skills from human video - Engadget · 2026-11-10
NVIDIA DLSS 4.5 Has Presets You Should Know About
This story is an original summary drafted with AI by Lazyfounder from the reporting listed above and checked by automated validation. Facts are attributed to their original publishers; sections marked as analysis are Lazyfounder's. Where a source is in another language, facts were machine-translated and quotations are reported, not reproduced. Read the original coverage via the links, and see our AI policy and corrections policy.
About the author
Editor, Lazyfounder
Tarun Mottlia edits LazyFounders, covering Indian startups, funding rounds, AI and product launches. Every story on the site is AI-assisted and checked against its cited sources before publication.
More stories by Tarun MottliaGet the LazyFounder Brief
Startup, funding and AI news in a five-minute read. Join the early-access list.
